You may get signed release tarballs from Yubico’s release page.
This project uses 'autoconf', 'automake', 'pkg-config' and 'libtool' to achieve portability and ease of use.
In addition, 'libfido2' (>=
1.3.0) is needed. Versions of this project up to 1.0.8 used
libu2f-host
and libu2f-server
. On Ubuntu, the necessary dependencies can be
installed using
# apt install --no-install-recommends autoconf automake libtool pkg-config \
libfido2-dev libpam-dev libssl-dev
If you downloaded a tarball, build it as follows.
$ ./configure
$ make
You may check out the sources using Git with the following command:
$ git clone https://github.com/Yubico/pam-u2f.git
This will create a directory 'pam-u2f'. Enter the directory:
$ cd pam-u2f
autoconf
, automake
, libtool
, and libpam
must be installed.
AsciiDoc
and xsltproc
are used to generate the manpages. On Ubuntu,
the necessary dependencies can be installed using
# apt install --no-install-recommends autoconf automake libtool \
pkg-config libfido2-dev libpam-dev libssl-dev asciidoc xsltproc \
libxml2-utils docbook-xml
On Fedora, the necessary dependencies can be installed using
# dnf -y install autoconf automake libtool libfido2-devel pam-devel \
openssl-devel asciidoc
Generate the build system using:
$ autoreconf --install
Then build as usual, see above under Building.
Once the module is built, copy the file pam_u2f.so
to the correct
directory for your system. Typically /lib/security/
or
/lib/x86_64-linux-gnu/security/
. This is automated by make install
assuming that the pam directory chosen by configure
is correct. If
that is not the case it can be specified with ./configure
--with-pam-dir=
.
Create a file for a new service in /etc/pam.d/
or edit an already
existing one by adding a line similar to this:
auth sufficient pam_u2f.so debug
For more information about the syntax of PAM configuration files, please see the manual page for pam.conf(5). Additional example configurations can be found below.
Important
|
An erroneous PAM configuration may lock some or all users out of the system or prevent you from gaining root privileges. It is recommended that you start a separate shell with root privileges while configuring PAM to be able to revert changes if something goes wrong. Test your configuration thoroughly before closing the root shell. |
debug |
Enables debug output |
debug_file |
Filename to write debugging messages to. If this file is missing, nothing will be logged. This regular file has to be created by the user or must exist and be a regular file for anything getting logged to it. It is not created by pam-u2f on purpose (for security considerations). This filename may be alternatively set to "stderr" (default), "stdout", or "syslog". |
origin=origin |
Set the relying party ID for the FIDO authentication procedure. If no value is specified, the identifier "pam://$HOSTNAME" is used. |
appid=appid |
Set the application ID for the FIDO authentication procedure. If no value is specified, the same value used for origin is taken ("pam://$HOSTNAME" if also origin is not specified). This setting is only applicable for U2F credentials created with pamu2fcfg versions v1.0.8 or earlier. Note that on v1.1.0 and v1.1.1 of pam-u2f, handling of this setting was temporarily broken if the value was not the same as the value of origin. |
authfile=file |
Set the location of the file that holds the mappings of user names to keyHandles and user keys. An individual (per user) file may be configured relative to the users' home dirs, e.g. ".ssh/u2f_keys". If not specified, the location defaults to $XDG_CONFIG_HOME/Yubico/u2f_keys. If $XDG_CONFIG_HOME is not set, $HOME/.config/Yubico/u2f_keys is used. |
expand |
Enables variable expansion within the authfile path: |
authpending_file=file |
Set the location of the file that is used for touch request
notifications. This file will be opened when pam-u2f starts waiting
for a user to touch the device, and will be closed when it no longer
waits for a touch. Use inotify to listen on these events, or a more
high-level tool like
yubikey-touch-detector.
Note that yubikey-touch-detector v1.5.1 and later no longer rely on the
authpending file for its detection mechanism. Set an empty value in order to
disable this functionality, like so: |
nouserok |
Set to allow authentication attempts to continue to other authentication
methods even if the user trying to authenticate is not found inside
|
openasuser |
Setuid to the authenticating user when opening the authfile. Useful
when the user’s home is stored on an NFS volume mounted with the
|
alwaysok |
Set to enable all authentication attempts to succeed (aka presentation mode). |
max_devices=n_devices |
Maximum number of devices (credentials) allowed per user (default is 24). Devices specified in the authorization mapping file that exceed this value will be ignored. |
interactive |
Set to prompt a message and wait before testing the presence of a FIDO device. Recommended if your device doesn’t have a tactile trigger. |
interactivenodevice |
Set to prompt a message and wait if no FIDO device is detected. Avoids silent authentication failures at the cost of notifying attackers fido devices are used for authentication |
[prompt=your prompt here] |
Set individual prompt message for interactive mode. Watch the square brackets around this parameter to get spaces correctly recognized by PAM. |
manual |
Set to drop to a manual console where challenges are printed on screen and response read from standard input. Useful for debugging and SSH sessions without U2F-support from the SSH client/server. If enabled, interactive mode becomes redundant and has no effect. |
cue |
Set to prompt a message to remind to touch the device. |
[cue_prompt=your prompt here] |
Set individual prompt message for the cue option. Watch the square brackets around this parameter to get spaces correctly recognized by PAM. |
nodetect |
Set to skip detecting if a suitable FIDO token is inserted before
performing the full tactile authentication. This detection was created
to avoid emitting the "cue" message if no suitable token exists,
because doing so leaks information about the authentication stack if a
token is inserted but not configured for the authenticating user.
However, it was found that versions of |
userpresence=int |
If 1, request user presence during authentication. If 0, do not request user presence during authentication. If omitted, fallback to the authenticator’s default behaviour. |
userverification=int |
If 1, request user verification during authentication (e.g. biometrics). If 0, do not request user verification during authentication. If omitted, fallback to the authenticator’s default behaviour. If enabled, an authenticator with support for FIDO2 user verification is required. |
pinverification=int |
If 1, request PIN verification during authentication. If 0, do not request PIN verification during authentication. If omitted, fallback to the authenticator’s default behaviour. If enabled, an authenticator with support for a FIDO2 PIN is required. |
sshformat |
Use credentials produced by versions of OpenSSH that have support for FIDO devices. It is not possible to mix native credentials and SSH credentials. Once this option is enabled all credentials will be parsed as SSH. |
Important
|
On dynamic networks (e.g. where hostnames are set by DHCP), users should not rely on the default origin and appid ("pam://$HOSTNAME") but set those parameters explicitly to the same value. |
Configure pam-u2f as a required
module after your primary authentication
module(s) for use as a second factor. Make sure that the primary authentication
method is not sufficient
or uses other control values that may preempt
execution of pam-u2f.
# Primary authentication method(s) above this line. auth required pam_u2f.so authfile=/etc/u2f_mappings cue
For a passwordless experience, where the authenticator PIN can be used in place of the user password, you can insert the below line towards the top of your service configuration. You may need to change the control value to something else if you’d like to execute additional authentication modules after pam-u2f.
auth sufficient pam_u2f.so authfile=/etc/u2f_mappings cue pinverification=1 # Fallback modules go below this line (e.g. password authentication).
Similar to the previous configuration but capable of built-in user verification, e.g. fingerprint matching using the YubiKey Bio. This example falls back to using PIN verification if the fingerprint does not match or is otherwise blocked.
auth sufficient pam_u2f.so authfile=/etc/u2f_mappings cue pinverification=0 userverification=1 auth sufficient pam_u2f.so authfile=/etc/u2f_mappings cue pinverification=1 userverification=0 # More fallback modules go below this line (e.g. password authentication).
A mapping must be made between the YubiKey token and the user name, see here for details on how to perform the registration using the bundled tool.
There are two ways to do this, either centrally in one file, or individually, where users can create the mapping in their home directories. If the central authorization mapping file is being used, user home directory mappings will not be used and the opposite applies if user home directory mappings are being used, the central authorization mappings file will not be used.
By default the mapping file inside a home directory will be opened as
the target user, whereas the central file will be opened as root
. If
the XDG_CONFIG_HOME
variable is set, privileges will not be dropped
unless the openasuser
configuration setting is set.
Important
|
Using pam-u2f to secure the login to a computer while storing the mapping file in an encrypted home directory, will result in the impossibility of logging into the system. The partition is decrypted after login and the mapping file can not be accessed. |
Create a file e.g. /etc/u2f_mappings
. The file must contain a user
name, and the information obtained during the registration procedure.
The mappings should look like this, one per line:
<username1>:<KeyHandle1>,<UserKey1>,<CoseType1>,<Options1>:<KeyHandle2>,<UserKey2>,<CoseType2>,<Options2>:... <username2>:<KeyHandle1>,<UserKey1>,<CoseType1>,<Options1>:<KeyHandle2>,<UserKey2>,<CoseType2>,<Options2>:...
Now add authfile=/etc/u2f_mappings
to your PAM configuration line,
so it looks like:
auth sufficient pam_u2f.so authfile=/etc/u2f_mappings
If you do not set the openasuser
setting, the authfile will be
opened and parsed as root
so make sure it has the correct owner and
permissions set.
Each user creates a .config/Yubico/u2f_keys
(default) file inside
their home directory and places the mapping in that file. You may want
to specify a different per-user file (relative to the users' home
directory), i.e. .ssh/u2f_keys
. Bear in mind, setting an absolute path
here is possible although very likely a fragile setup, and probably
not exhibiting the intended behaviour.
The file must have only one line:
<username>:<KeyHandle1>,<UserKey1>,<CoseType1>,<Options1>:<KeyHandle2>,<UserKey2>,<CoseType2>,<Options2>:...
This is much the same concept as the SSH authorized_keys file.
In this case, pam-u2f will drop privileges and read the mapping file
as that user. This happens regardless of the openasuser
option being
set.
Note that if you set the XDG_CONFIG_HOME
variable, privileges will not
be dropped by default. Consider also setting openasuser
in that
case.
In order to obtain the required information for the authentication procedure, a token should be first registered. This can be done by using the command line configuration tool provided with the module:
$ pamu2fcfg -uusername -opam://myorigin -ipam://myappid
the tool will register a connected token by using the specified origin
and appid. If neither are specified they will default to
pam://$HOSTNAME
. Additionally, it is possible to set other options
to require user presence (touch), PIN verification and resident
credentials. On success the tool prints to standard output a
configuration line that can be directly used with the module. For
additional information on the tool read the relative manpage (man
pamu2fcfg
).
For authenticator management (e.g. setting a PIN, enrolling fingerprints, and
more), please refer to
fido2-token
,
yubikey-manager
, or some other
suitable tool.
To generate SSH credentials OpenSSH version 8.2 or later is required. It is then possible to generate a credential file with:
$ ssh-keygen -t ecdsa-sk -f ./credential.ssh
Supported key types are ecdsa-sk and ed25519-sk. Note that passphrase protected credentials are currently not supported.
To use this credential the authfile
parameter should be set to the path of
the file credential.ssh
and the sshformat
option should also be set. If the
authfile
parameter is not set, it defaults to ~/.ssh/id_ecdsa_sk
.
Multiple devices (credentials) are supported. If more than one credential is specified, authentication against them is attempted sequentially as they are defined in the authorization mapping file. If during an authentication attempt a connected device is removed or a new device is plugged in, the authentication restarts from the top of the list.
Due to an issue with Fedora Linux, and possibly with other
distributions that use SELinux, a system configured with pam-u2f may
end up in a situation where access to the credentials file is denied.
If the nouserok
option is also set, this will result in a successful
authentication within the module, without using the FIDO
authenticator.
In order to correctly update the security context the command
fixfiles onboot
should be used on existing installations
Moreover, to allow read access to an authfile or directory placed in a non-standard location, the command
# chcon -R -t auth_home_t /path/to/target
should be used.
For more information see HERE.
Devices that solely support FIDO U2F and not FIDO2, e.g. the YubiKey 4 series,
can be used only in conjunction with compatible features. Enabling incompatible
features, such as setting the +pin
or the +verification
flags in the
authfile
or the corresponding options in the PAM service configuration causes
the device to be ignored.